BoE and FCA published a report on the results of a joint survey by BoE and FCA in 2019 to better understand the use of machine learning in the financial services sector in UK. A total of 106 responses were received from this survey, which was sent to almost 300 firms, including banks, credit brokers, e-money institutions, financial market infrastructure firms, investment managers, insurers, non-bank lenders, and principal trading firms. With respect to the respondent firms, the joint BoE-FCA report presents a quantitative overview of the use of machine learning, the machine learning implementation strategies of firms, approaches to the governance of machine learning, and the share of applications developed by third-party providers. Also covered are the perceptions of risks and ethical considerations and perspectives on constraints to development and deployment of machine learning, along with a snapshot of the use of different methods, data, safeguards performance metrics, validation techniques, and perceived levels of complexity.
The survey collected information about the nature of deployment of machine learning, the business areas where it is used, and the maturity of applications. It also collected information on the technical characteristics of specific machine learning use cases. This included how the models were tested and validated, the safeguards built into the software, the types of data and methods used, and considerations around benefits, risks, complexity, and governance. The key findings of the survey include the following:
- Machine learning is increasingly being used in UK financial services. Two-third of the respondents reported that they already use it in some form. The median firm uses live machine learning applications in two business areas and this is expected to more than double within the next three years.
- In many cases, machine learning development has passed the initial development phase and is entering more mature stages of deployment. One-third of the machine learning applications are used for a considerable share of activities in a specific business area. Deployment is most advanced in the banking and insurance sectors.
- Regulation is not seen as an unjustified barrier but some firms stress the need for additional guidance on how to interpret the current regulation. The biggest reported constraints are internal to firms, such as legacy IT systems and data limitations.
- Firms thought that machine learning does not necessarily create new risks, but could be an amplifier of the existing ones. Risks, such as machine learning applications not working as intended, may occur if model validation and governance frameworks do not keep pace with technological developments.
- Firms validate machine learning applications before and after deployment. The most common validation methods are outcome-focused monitoring and testing against benchmarks. However, many firms note that machine learning validation frameworks still need to evolve in line with the nature, scale, and complexity of machine learning applications.
- Firms mostly design and develop machine learning applications in-house. However, they sometimes rely on third-party providers for the underlying platforms and infrastructure, such as cloud computing.
- The majority of users apply their existing model risk management framework to machine learning applications. However, many highlight that these frameworks might have to evolve in line with the increasing maturity and sophistication of machine learning techniques.
This survey was the first step toward better understanding the impact of machine learning on UK financial services and forms the basis for a conversation about how safe machine learning deployment can be supported going forward. To foster further conversation around machine learning innovation, BoE and FCA have announced plans to establish a public-private group to explore some of the questions and technical areas covered in this report. BoE and FCA are also considering repeating this survey in 2020.
Keywords: Europe, UK, Banking, Insurance, Securities, Machine Learning, Governance, Risk Management, Fintech, FCA, BoE
Previous ArticleBCBS Report Examines Progress on Adoption of Basel Framework
EIOPA submitted—to the European Parliament, the Council of the European Union, and EC—its 2020, fifth, and last annual report on long-term guarantee measures and measures on equity risk.
The BIS Innovation Hub Swiss Centre, SNB, and the financial infrastructure operator SIX announced the successful completion of a joint proof-of-concept (PoC) experiment as part of the Project Helvetia.
EBA published the final draft regulatory technical standards for calculation of own funds requirements for market risk, under the standardized and internal model approaches of the Fundamental Review of the Trading Book (FRTB) framework.
EIOPA published discussion paper on a methodology for the potential inclusion of climate change in the Solvency II (sometimes also written as SII) standard formula when calculating natural catastrophe underwriting risk.
EU published, in the Official Journal of the European Union, corrigenda to the Directive and the Regulation on the prudential requirements and supervision of investment firms.
MAS proposed amendments to certain regulations, notices, and guidelines arising from the Banking (Amendment) Act 2020.
PRA published a statement that explains when to expect further information on the PRA approach to transposing the Capital Requirements Directive (CRD5), including its approach to revisions to the definition of capital for Pillar 2A.
RBNZ launched consultations on the scope of the Insurance Prudential Supervision Act (IPSA) 2010 and on the associated Insurance Solvency Standards.
SRB published the work program for 2021-2023, setting out a roadmap to further operationalize the Single Resolution Fund and to achieve robust resolvability of banks under its remit over the next three years.
EIOPA is consulting on the relevant ratios to be mandatorily disclosed by insurers and reinsurers falling within the scope of the Non-Financial Reporting Directive as well as on the methodologies to build these ratios.